Data Analysis and Data Mining by Scarpa Bruno Azzalini Adelchi & Bruno Scarpa

Author:Scarpa, Bruno, Azzalini, Adelchi & Bruno Scarpa
Language: eng
Format: epub
Publisher: Oxford University Press, USA
Published: 2012-11-15T00:00:00+00:00

Figure 4.30 Insurance customers: profiles of lasso coefficients as tuning parameter s is varied. Standardized coefficients plotted versus

Lasso shrinks parameters and gives a model with only 9 variables (15 parameters). The estimated coefficients obtained by lasso with R are listed in table 4.7.

We also fit some nonlinear models to the data. To choose a suitable neural network for our problem, we divide the training set into two subsets of equal size and fit a number of different networks on the first subset by modifying the number of nodes in the hidden layer and the weight decay. Networks with 10 to 19 hidden nodes and 10 values for weight decay between 0.001 and 0.1 are evaluated, and we select the one with the smallest squared prediction error on the second subset of the training set. The best model has 12 nodes and weight decay of 0.1.

Table 4.7. INSURANCE CUSTOMERS:ESTIMATE OF COEFFICIENTS FOR BEST LASSO NONZERO ESTIMATE OF LINEAR MODEL

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.